Gigantic negative magnetoresistance in a disordered topological insulator

TL;DR

This paper reports a giant negative magnetoresistance of 98% in a disordered topological insulator, revealing a new disorder-related mechanism supported by simulations, distinct from magnetic or chiral anomaly effects.

Contribution

It introduces a novel disorder-induced mechanism for large negative MR in topological insulators, expanding understanding beyond magnetic and chiral anomaly origins.

Findings

Gigantic negative MR of 98% at 14 T and 10 K in TlBi0.15Sb0.85Te2

Numerical simulations suggest Zeeman effect on disordered current paths as the cause

Disorder engineering can induce large negative MR in narrow-gap systems

Abstract

With the recent discovery of Weyl semimetals, the phenomenon of negative magnetoresistance (MR) is attracting renewed interest. While small negative MR can occur due to the suppression of spin scattering or weak localization, large negative MR is rare in materials, and when it happens, it is usually related to magnetism. The large negative MR in Weyl semimetals is peculiar in that it is unrelated to magnetism and comes from chiral anomaly. Here we report that there is a new mechanism for large negative MR which is not related to magnetism but is related to disorder. In the newly-synthesized bulk-insulating topological insulator TlBi$_{0.15}$Sb$_{0.85}$Te$_2$, we observed gigantic negative MR reaching 98% in 14 T at 10 K, which is unprecedented in a nonmagnetic system. Supported by numerical simulations, we argue that this phenomenon is likely due to the Zeeman effect on a barely percolating current path formed in the disordered bulk. Since disorder can also lead to non-saturating linear MR in Ag$_{2+\delta}$Se, the present finding suggests that disorder engineering in narrow-gap systems is useful for realizing gigantic MR in both positive and negative directions.

Equations

$$n$$

$$=\frac{1}{3{\pi }^2{\hslash }^3}{\left(\frac{2m^{\ast }{e}^2}{4\pi ϵ{ϵ}_0}\right)}^{3/2}{N}^{1/2}.$$

$$H=\frac{\Delta }{2}\sum _i{f}_i{n}_i+\frac{1}{2}\sum _{i\ne j}{V}_{ij}{q}_i{q}_j.$$

$$\Delta \sigma =\frac{e^2}{2{\pi }^2\hslash }{f}_3\left(\frac{B}{B_{in}}\right){\left(\frac{eB}{\hslash }\right)}^{1/2},$$

Full text

Magnetic fields tend to localize electrons by trapping them in cyclotron orbits, leading to positive magnetoresistance (MR) in metals and semiconductors. Large negative MR is therefore unusual and is always a signature of some peculiar physics. So far, three mechanisms are widely known to cause large negative MR: (i) quenching of the Kondo effect Andrei1983 ; Hanaki2001 , (ii) magnetic-field-induced phase transition from a paramagnetic insulator to a ferromagnetic metal (so-called colossal MR) RamirezCMR , and (iii) chiral anomaly in Weyl semimetals NIELSEN1983 ; Aji2012 ; Son2013 ; Burkov2014 . The last one is expected to be observed only in the longitudinal configuration (i.e. when the current is parallel to the magnetic field) and has been of significant interest since three-dimensional (3D) Dirac semimetals and Weyl semimetals started to attract attentions Wan2011 ; Young2012 . Intriguingly, large negative, longitudinal MR of similar character has been observed in materials having no Weyl nodes Kikugawa2016 , and the meaning of negative MR in nonmagnetic systems is currently being scrutinized Goswami2015 ; Arnold2016 . In this context, finding of a new mechanism for large negative MR in zero-gap or narrow-gap systems without magnetism would be important for establishing general understanding of the negative MR. This paper reports an unexpected discovery of gigantic negative MR in a nearly-bulk-insulating topological insulator (TI), which turned out to be a platform to realize large negative MR in the bulk through a new mechanism related to disorder.

The key feature of 3D TIs is the topologically-protected surface states Hasan2010 ; Ando2013 , and significant efforts have been devoted to find ways to achieve sufficiently bulk-insulating samples to address the transport properties through the surface states Ando2013 ; Ren2010 . The most successful strategy to achieve this goal has been to adopt the concept of compensation, namely, balancing the numbers of donors and acceptors Ren2011 . This has led to the successful achievements of surface-dominated transport in bulk single crystals of Bi2-xSbxTe3-ySey (BSTS) Taskin2011 . We have recently synthesized another solid-solution system of 3D-TIs, TlBixSb1-xTe2 (TBST), which belongs to the family of thallium-based ternary III-V-VI2 chalcogenides Lin2010 ; Yan2010 ; Eremeev2011 ; Sato2010 ; Chen2010 and the chemical potential can be tuned through the bulk band gap between $n$- and $p$-type states Trang2016 . We found that the maximal compensation in TBST is achieved near $x\simeq$ 0.2 (see Supplementary Information for details), where one can obtain nearly bulk-insulating samples. The gigantic MR is discovered in such a compensated sample of TBST.

I Results

Figure  1 shows the temperature dependencies of $\rho_{xx}$ in TBST for various $x$ values. Among the samples showing metallic behavior, $x$ = 0 and 0.1 are $p$-type, while $x$ = 0.6 is $n$-type [see Figs. 2a and 2c for the Hall data]. A slight increase in $x$ from 0.1 to 0.15 causes $\rho_{xx}$ at $\sim$10 K to increase by three orders of magnitude to 0.4 $\Omega$cm, and an insulating behavior ($\partial\rho_{xx}/\partial T<0$) arises between 10 to 100 K. We focus on this $x=0.15$ sample, in which the Hall carrier density $n_{\mathrm{H}}$ extracted from the zero-field slope of $\rho_{yx}(B)$ (Fig. 2b) is only 1.9 $\times$ 1015 cm*-3* at 2 K and approximately follows the Arrhenius law with an activation energy $\Delta_{\mu}$ of 13 meV [Fig. 2d]. The compensation in TBST is not as effective as that in BSTS Taskin2011 and the maximum attainable resistivity is lower; consequently, we found no evidence for a surface contribution to the transport properties.

The surprising finding is that the magnetic field of 14 T leads to a reduction of $\rho_{xx}$ at 10 K by nearly two decades from 400 to 6.5 m$\Omega$cm (orange line in Fig. 1). How this gigantic negative MR evolves with temperature is shown in Fig. 3a; with decreasing temperature, the MR changes from entirely positive at $\gtrsim 200$ K to entirely negative at $\lesssim 20$ K. The additional cusp-like feature seen at 2 K for $B\lesssim$ 1 T may be attributed to a weak localization effect, but it cannot explain the large negative MR; this can be seen by the estimate of the possible 3D weak localization contribution Altshuler for 2 T shown in Fig. 3a (see Supplementary Information for details). The logarithmic plot of $\rho_{xx}(B)$ shown in Fig. 3b suggests that the MR behavior approaches a $1/B^{2}$ dependence in high fields at low temperature.

It is important to note that the compositional analysis using inductively-coupled plasma atomic-emission spectroscopy found no magnetic impurities in TBST, which is corroborated by the magnetic susceptibility data (inset of Fig. 1). Hence, magnetism is not involved in this gigantic negative MR. Also, the chiral anomaly is irrelevant, because the negative MR shows up in the transverse configuration. Its origin is probably not related to the anisotropy in the band structure, because the difference between $B\parallel c$ and $B\perp c$ is negligible (Fig. 3c).

Phenomenologically, the appearance of the negative MR in TBST is tied to the emergence of an insulating behavior. In fact, among the investigated samples, all those that show an insulating behavior ($0.13\leq x\leq 0.4$) presented a negative MR which was most pronounced in the $x$ = 0.15 sample shown here (additional data for other $x$ values are shown in the Supplementary Information). The onset temperature of the negative MR upon cooling compares well with the onset of an activated behavior in $\rho_{xx}(T)$. A fit of the data below 100 K to $\rho_{xx}\propto e^{E_{A}/k_{\mathrm{B}}T}$, shown as a dashed green line in Fig. 1, yields an activation energy $E_{A}\approx$ 6 meV. This value is substantially smaller than the bulk band gap of $\sim$0.1 eV Chen2010 ; Trang2016 . Such a small effective activation energy in compensated TIs Ren2010 ; Ren2011 has been discussed to be a signature of the formation of charge puddles Chen2013 ; Skinner2013 ; Skinner2012 , which we discuss in the next section to be responsible for the gigantic negative MR.

II Discussions

**Electron puddles. ** The concept of electron and hole puddle formation has been adopted in the framework of compensated TIs to explain their small effective activation energy Skinner2012 ; Skinner2013 ; Chen2013 as well as the unusual temperature dependence of the bulk optical conductivity Borgwardt2015 . Basically, the strong fluctuations of the Coulomb potential introduced by positively charged empty donors and negatively charged occupied acceptors, which coexist in compensated systems, inevitably create regions where the conduction-band bottom or the valence-band top cross the chemical potential, leading to the appearance of locally conducting puddles containing either electrons or holes Shklovskii1972 .

Similar to other compensated TIs, the impurity states in TBST are shallow due to a rather large dielectric constant $\epsilon\approx 200$, approximated from that of TlSbTe2 Deger2015 . Thus, their energies are close to the edge of the conduction or valence band. The $x$ = 0.15 sample is $n$-type and is slightly away from complete compensation, leading to the pinning of the chemical potential at low to moderate temperature to the donor impurity levels, which lie below but close to the conduction band bottom. The obtained activation energy $\Delta_{\mu}$ = 13 meV corresponds directly to $E_{0}-\mu$, which is the energy difference between the conduction band bottom in 0 T, $E_{0}$, and the chemical potential $\mu$. The dominance of a single $n$-type channel at moderate temperature is also indicated by the linear $\rho_{yx}(B)$ behavior at $T\geq$ 40 K (Fig. 2e).

At low temperature where the screening due to thermally activated carriers is negligible, the fluctuations of the Coulomb potential arising from charged impurities pile up. In this situation, the same mechanism as in the case of electron-hole puddles Skinner2012 enforces the formation of spatially separated electron puddles. Skinner, Chen, and Shklovskii Skinner2012 pointed out that one obtains effectively a reduced activation energy in the presence of strong fluctuations of the Coulomb potentials. Hence, the activation energy $E_{A}$ ($\approx$ 6 meV) observed in the low-temperature $\rho_{xx}(T)$ data is actually expected to differ from $\Delta_{\mu}$ extracted from the $n_{\rm H}(T)$ behavior at higher temperature.

Below 4 K, the $\rho_{xx}(T)$ behavior for $x$ = 0.15 is metallic; this is different from the prediction for a compensated TI in Ref. Skinner2012 , where a variable range hopping transport between puddles was postulated. Furthermore, in the case of hopping conduction, a positive MR would be expected due to the decreasing size of the atomic orbitals in strong magnetic fields Shklovskii1984 ; ESbook . The metallic behavior indicates that either the electron puddles in the slightly doped samples weakly percolate or another conduction mechanism (e.g., via impurity bands) connects them.

**Numerical Simulations. ** Based on the above picture of the spatially disordered electronic states caused by electron puddles, we propose that the increasing conductivity in applied magnetic fields is a result of an increase of percolating puddles due to the Zeeman energy. To understand this effect, we performed numerical simulations of an imperfectly compensated Coulomb system. The characteristic energy scale of such a system is given by the Coulomb energy between neighboring dopants $E_{c}=e^{2}/(4\pi\epsilon\epsilon_{0}N^{-1/3})$ Borgwardt2015 , where $N$ denotes the dopant density. Since the dielectric constant in TBST is large ($\epsilon\approx 200$ Deger2015 ) as discussed above, the temperature scale $E_{c}/k_{B}$ becomes small and we use this parameter to adjust the numerical results to the experimental data. Also, as the negative MR is most pronounced at low temperature, we restrict the analysis to the case $k_{B}T\ll E_{c}$. For the simulation, the degree of compensation was set at $K=N_{A}/N_{D}=0.9-0.95$, where $N_{A}$ and $N_{D}$ denote the densities of acceptors and donors, respectively. This implies that the chemical potential is close to the conduction band edge and that puddles are almost exclusively formed by electrons. The electron puddles are spatially confined to regions where the position-dependent electrostatic potential $\phi({\mathbb{r}})$ (which is responsible for the local band bending) exceeds $E_{0}-\mu$; in other words, electron puddles appear when the local band bending $\phi({\mathbb{r}})$ brings down the conduction band edge below the chemical potential.

Assuming a single spherical band with effective mass $m^{*}$, the typical density of charges delocalized within the puddles at low temperature is given by

[TABLE]

For $\epsilon=200,m^{*}=0.2m_{e}$ and $N=3\times 10^{20}\,\mathrm{cm}^{-3}$, we find a density $n\approx 10^{17}\,\mathrm{cm}^{-3}$, which is significantly smaller than that of the dopants, i. e. $n\ll N$. Thus, the impurity states serve as a reservoir for the delocalized states and approximately fix the chemical potential.

In such a situation, the volume accessible by the delocalized states (i.e. the volume occupied by the puddles) increases with increasing magnetic field $B$ due to the Zeeman band shift. Adopting $g\approx 6$ from a similar system TlBiSSe Novak2015 , the down shift of the band bottom, $g\mu_{\mathrm{B}}B$, is estimated to be $0.35\,\mathrm{meV}/\mathrm{T}$. This effect is visualized in Fig. 4, which shows a typical simulation result for $K=0.95$ (details are explained in the Methods section). Here we plot the surface contours where the local electrostatic potential supports puddle formation, i.e. where $\phi(\mathbb{r})=(E_{0}-\mu)-E_{\rm Z}(B)$ is satisfied [$E_{\rm Z}(B)=g\mu_{\rm B}B$ is the Zeeman energy], for $E_{\rm Z}/E_{c}$ = 0, 0.5, and 1, corresponding to $B\simeq$ 0, 7, and 14 T, respectively. Note that, due to the pinning of the chemical potential to the impurity levels, $E_{0}-\mu$ is expected to be unchanged with $B$. Within the enclosed regions (i.e. electron puddles), delocalized states exist. One can clearly see that with increasing magnetic field, previously disconnected puddles merge. Note that the occupation of most impurity states, and therefore the profile of the potential fluctuations, is assumed not to change with magnetic field. The increase in the percolating paths in $B>0$ demonstrated in Fig. 4 naturally leads to an increase in the electrical conductivity.

To quantify this effect we have calculated the conductivity at different magnetic fields using a random resistor network Derrida1984 (details are explained in the Methods section). The local conductivity of each resistor in this network was calculated according to $\sigma(\mathbf{r})=\mu_{e}n(\mathbf{r})$, where we have assumed a constant electron mobility $\mu_{e}$. For the calculation of the density $n(\mathbf{r})$ we have assumed a locally free electron gas which gives $n(\mathbf{r})\sim\left(\phi(\mathbb{r})-(E_{0}-\mu)+E_{\rm Z}(B)\right)^{3/2}$ and $\sigma(\mathbf{r})=0$ if $\phi(\mathbb{r})<(E_{0}-\mu)-E_{\rm Z}(B)$. Note that this approach is tailored to describe the regime where puddles overlap and start to percolate. In this regime, variable-range hopping effects Skinner2012 (not included in our resistor network) can be neglected. Furthermore, weak localization effects, which we have already shown to be small, are not included. By fitting the numerical results at $B=0$ to the experimental data, we find a mobility of $\mu_{e}\approx 11,500\,\mathrm{cm}^{2}$/Vs corresponding to a scattering rate of $1/\tau\approx 10^{12}/$s, which is in reasonable agreement with the scattering rates measured in similar systems, see Ref. Borgwardt2015 for an overview. These numbers depend, however, sensitively on $K$ and therefore are only an order of magnitude estimate. Furthermore, one should note that this number only characterizes the typical mobility within a puddle, not the effective mobility obtained by averaging over the bulk. The shape and width of $\rho_{xx}(B)/\rho_{xx}(B=0)$ depend only weakly on $K$, while $E_{c}$ determines the width of the curve and is used as an adjusting parameter; we find $E_{c}=60\,\pm 10$ K reproduces the data well for $g=6$. We show in Figs. 3a and 3b our numerical results for $\rho_{xx}(B)/\rho_{xx}(B=0)$ with $K=0.9$ and $E_{c}=56\,$K, which correspond to a dopant densisty $N=3\times 10^{20}\,\mathrm{cm}^{-3}$ and is in reasonable agreement with studies of similar systems Borgwardt2015 . Thus, a set of reasonable parameters reproduces the observed negative MR in our random resistor network simulation. Remarkably, our numerics even reproduces the approximate $1/B^{2}$ asymptotics of $\rho_{xx}$ for high fields, see Fig. 3b. In even higher fields, when either a large volume fraction of the sample becomes a good metal or when our approximation of a fixed chemical potential breaks down, we expect a modified high-field asymptotics.

**Implications. ** The above mechanism for the MR is expected to be independent of whether the configuration is transverse or longitudinal. Indeed, as is shown in the Supplementary Information, essentially the same negative MR is observed in the longitudinal configuration. As for the temperature dependence, although our numerical calculations were performed only for 0 K, it is naturally expected that the random Coulomb potential is gradually screened with increasing temperature by thermally-activated carriers, which results in smearing of the puddles; consequently, the negative MR is gradually diminished. The positive MR observed at high temperature is relatively large and would be an interesting topic for future studies.

It is prudent to mention that large negative MR has not been observed in other compensated TI materials such as BSTS. Our numerical simulations show that for the occurrence of the large negative MR, it is important that the parameter $K$ is at a right value away from 1.0, which means that the compensation is not perfect and the puddles of only one carrier type is weakly percolating in 0 T. Obviously, such a situation is not realized in BSTS, in which the degree of compensation is better than that in the present TBST series. We also note that the density of naturally created vacancies in the crystals may widely vary among different compounds, resulting in significantly different landscape/density of puddles and hence different percolation behavior.

As our simulations suggest, the gigantic negative MR in TlBi0.15Sb0.85Te2 is likely due to a new mechanism related to the spatial disorder of the electronic system in the form of puddles, which results in magnetic-field-sensitive percolation of the current paths. In this regard, it is interesting that large positive, non-saturating linear MR has been shown to originate from spatially distorted current paths in Ag2+δTe and Ag2+δSe Parish2003 , which are narrow-gap semiconductors and are potentially TIs WZhang2012 ; such distorted current paths could lead to negative MR, but only for the longitudinal configuration Hu2007 . Obviously, disorder in narrow-gap systems plays a key role in both positive and negative gigantic MR, and its understanding may help to engineer practical devices for magnetic data storage and sensors.

III Methods

Sample preparations. Single crystals of TlBixSb1-xTe2 were grown from a melt of high-purity (99.9999%) starting materials of elementary Tl, Bi, Sb and Te shots that were cleaned to remove the oxide layers formed in air Wang2015 . They were mixed in the target composition and sealed in evacuated quartz tubes. After heating to 600∘C for 48 h, at which the tubes were intermittently shaken to ensure homogeneity of the melt, they were slowly cooled down to 400∘C in 100 h. Typically the growth resulted in multiple intertwined crystallites whose size increases from bottom of the grown boule to the top, and single crystals suitable for further analysis were prepared by cleavage at the top part. The actual chemical composition of the samples was confirmed to be consistent with the nominal one by inductively-coupled plasma atomic-emission spectroscopy (ICP-AES) as well as by energy-dispersive X-ray spectroscopy (EDX). The crystal structure was confirmed by powder X-ray diffraction to be unchanged from both endmembers Hockings1961 over the whole range of $x$.

Measurements. Transport measurements were performed on thin needle-like samples with a typical dimension of 4$\times$1 mm2 in the $ab$ plane and 0.4 mm in the $c$ axis. Using a bath cryostat equipped with a variable-temperature insert the resistivity $\rho_{xx}$ and the Hall resistivity $\rho_{yx}$ were measured simultaneously in sweeping magnetic fields ($\pm$14 T) with a low-frequency lock-in technique. Electrical contacts were prepared in a standard six-probe configuration using 25 $\mu$m platinum wires attached with dissolved silver paint that was cured at room temperature.

Numerical simulations. To simulate the behavior of the puddles in a magnetic field we have used the common static model of shallow donors and acceptors in compensated semiconductors Skinner2012 ; Skinner2013 ; Borgwardt2015 . The donors with density $N_{D}$ and the acceptors with denity $N_{D}=KN_{A}$ are placed at random, uncorrelated positions. Their bare energies are seperated by a band gap $\Delta$ which was chosen as $15\times E_{c}$ in the simulations.

For a specific realization of the dopant positions (say $\mathbf{r}_{1},\mathbf{r}_{2},\dots$), the Hamiltonian reads

[TABLE]

$q_{i}$ denotes the charge in units of the elementary charge of the dopant at position $\mathbf{r}_{i}$. It can be either [math] or $-1$ for acceptors ($f_{i}=-1$) and either [math] or $+1$ for donors ($f_{i}=+1$). The occupation $n_{i}$ of the $i$-th dopant is related to its charge $q_{i}$ by $q_{i}=(f_{i}+1)/2-n_{i}$.

The Coulomb interaction $V_{ij}$ between the dopants is equipped with a short distance cutoff of the order of the effective Bohr radius $a_{B}=4\pi\varepsilon\varepsilon_{0}\hbar^{2}/(m^{*}e^{2})$, where $m^{*}$ denotes the effective mass. This accounts for the finite extent of the dopant wavefunctions Skinner2012 ; Skinner2013 . With this approximation the effective interaction reads $V_{ij}=\tfrac{e^{2}}{4\pi\varepsilon\varepsilon_{0}\sqrt{|\mathbf{r}_{i}-\mathbf{r}_{j}|^{2}+a_{B}^{2}}}$. The model is valid for $\Delta\gg E_{c}=e^{2}/(4\pi\varepsilon\varepsilon_{0}N^{-1/3})$ (otherwise one has to include the band states) and $\varepsilon\gg 1$ to ensure that the dopants are shallow.

We have performed simulations for zero temperature. Only small temperature effects are expected as long as $T\ll E_{c}$ Borgwardt2015 . To find the true groundstate is an exponentially hard problem, but there is an algorithm to find an approximate groundstate, called a pseudo-groundstate, in polynomial time ESbook ; Skinner2012 . The physical properties of a pseudo-groundstate are expected to be indistinguishable from that of the true groundstate.

We introduce the single electron energies as $\epsilon_{j}=\frac{\Delta}{2}f_{j}-\sum_{i\neq j}V_{ij}\;q_{i}$. A pseudo-groundstate is found when $\delta E_{(i,j)}=\epsilon_{j}-\epsilon_{i}-V_{ij}>0$ is satisfied for all proper pairs with $n_{j}=0$ and $n_{i}=1$. Simulations are performed in a cubic volume $V=L^{3}$ with periodic boundary conditions. Once we have found a pseudo-groundstate, we plot the surface contour of constant potential as explained in the main text, where a typical result is shown.

For the calculation of the conductivity we proceed as follows once a pseudo-groundstate is found: The simulation volume is discretized and the grid points serve as nodes for the resistor network. We calculate the local conductivity between two neighboring grid points from the Coulomb potential and the Zeeman energy as explained in the main text. The emerging resistor network is solved exactly, see Ref. Derrida1984 for details on the algorithm (however, in contrast to Ref. Derrida1984 , we use physical boundary conditions where the currents, rather than the potentials, are set to zero on the growing face of the network). In this way, the conductivity of the full system is found. The grid size was chosen as $N^{-1/3}$, which is the typical lengthscale ($\sim$1.5 nm for $N=3\times 10^{20}$ cm*-3*) on which the potential changes. By using finer grids, we have checked that discretization effects are small ($\lesssim 5\%$ and well within the numerical error bars). The simulations have been performed for $\approx 4.1\times 10^{5}$ dopants with periodic boundary conditions. The orbital MR inside the puddles is neglected in the present calculations.